There is a lot of great info in this thread and I may be repeating some of it here, so I apologize if this has already been covered. But I wanted to describe most of what I am planning for my DAC modification. I have not yet implemented this yet, so I can’t be totally confident that I haven’t made mistakes in my understanding.
My prior rush to build an active Vocm circuit was somewhat of a waste of time and a bit of money since I didn’t adequately investigate what was needed. The output buffers are driven by an OnSemi MC100E452FNG which is a high-speed 5-bit differential register operating on 5V PECL logic. This device generates an 800mV pk-pk signal ranging from 3.3V to 4.1V, so the best Vocm voltage is 3.7V (middle of this range) in order to limit wasted DC current through the multi-feedback reconstruction filters.
I am going to swap out the AD8132 op amps for AD8139 parts. These parts have considerably lower broadband noise, lower distortion, lower input offset voltage, and have a Vocm input that is easier to drive. But the Vocm input still benefits from a low impedance, low noise source from an active circuit (although the impact will be less than with the AD8132).
Keeping the two 1uF capacitors on each Vocm input is still beneficial since they lower the effective impedance of the Vocm input (particularly at higher frequencies) and reduce crosstalk between the parts. But they do complicate the active Vocm circuit. 8uF is considered a fairly large load for a high frequency op amp, and creates an unintended low frequency pole in the feedback loop which introduces a significant phase shift at the unity-gain crossover frequency that can cause the op amp to oscillate. To avoid this, a small series resistor must be added to the output of the op amp. To compensate for the voltage drop across this resistor, the feedback is taken after this resistor, and a small AC feedback resistor is added directly from the op amp output. Here is the circuit I came up with.
This circuit will be implemented on a small PCB in the same way that APS does their mod. The R5 and R6 resistors currently used to generate the Vocm voltage will be removed.
Changing to AD8139 op amps requires modifications to the power supply that powers these circuits since the current requirements are more than double and the input voltage range is more restricted. I plan to remove the two Q700 transistors and replace each with a small PCB regulator based on the LT3045. This is an ultra-low noise LDO (low dropout voltage) regulator with 0.8uV of output noise. I’m going to operate these regulators at 8V since this voltage allows the AD8139 op amps to operate at their lowest distortion with ample headroom above and below the voltage swing while keeping power dissipation reasonable. The estimated current requirements for the four AD8139 op amps and active Vocm circuit is about 125mA factoring in dynamic load currents - well within the capabilities of the LT3045 with a modest heatsink.
I plan to use a separate linear power supply to power the 12V input on the main PCB, so this frees up the 12V circuit on the existing power supply. The LT3045 is designed as a low dropout device, meaning that it will work effectively with an input voltage that is not much higher than the output. At the anticipated 125ma of current and 8V output, a 9V input supply should suffice, but I’m going to use 9.5V to have a little extra headroom. This keeps the regulator power dissipation to below 200mW. So the current 12V regulator on the existing power supply will be replaced with a regulator set to 9.5V. Since this regulator will only be supplying the 250ma of power for the output stage, the existing heatsinks can easily handle the increased voltage drop.
There are currently four LM317 regulators on the existing power supply (and Bridge interface) board. These regulators were first introduced 50 years ago and compared to modern regulators, they are fairly high noise with limited power supply rejection. If the Bridge is not installed, only two of these are actually used - one for the main 12V supply and one for the 5V logic.
I plan to replace the LM317 regulators with Sparkos discrete regulators.The Sparkos regulators are not as low-noise as the LT3045, but can handle more power and have excellent low-frequency PSRR so they do a great job of eliminating mains hum. I plan to allocate the regulators as follows:
- 9.5V as a pre-regulator to supply power to the LT3045s used for the output stage.
- 5V for the front display
- 5V as a pre-regulator for the oscillator power supply (more on this in a bit)
- 5V for the digital board and SD card circuit
I was planning to design my own power supply to power the 12V input on the main PCB, but I realized that I had an Ian Canada LinearPi Pro Mk ii that I wasn’t using and this will provide plenty of power at very low noise. This board uses the LT3042 (baby brother of the LT3045) along with a pass transistor to deliver up to 2A with 0.8uV of noise.
The other area that can benefit from power supply modifications is the Crystek oscillator. The Crystek is an excellent part, but is very susceptible to power supply noise. The current implementation uses multiple stages of filtering to reduce noise and a TI LMP7701 op amp acting as an error amplifier to control the voltage supplied to the oscillator. This is pretty good, but not as good as the LT3045, particularly at high frequencies (above 100khz). I am going to look into a dual LT3045 (or LT3042) circuit which reduces noise to 0.5uV (10Hz-100Khz). This regulator must be mounted very close (ideally < 1”) from the oscillator to minimize inductance and prevent noise pickup in the connections.
I plan to use balanced outputs only and will replace the board-mounted connectors with higher-quality chassis mount connectors. There may be benefit in hardwiring the transformer secondary outputs directly to the connectors, bypassing the existing circuitry. It looks like this circuitry provides muting and optional attenuation. Once I pull the existing transformers to replace them with the APS transformers, I’ll take a look into this.
